Re: [PATCH v2 3/5] media: sunxi: Add A10 CSI driver

Hi Maxime,

On Wed, Feb 06, 2019 at 10:16:05PM +0100, Maxime Ripard wrote:

Hi Sakari,

Thanks for your review, I have a few questions though, and the rest
will be addressed in the next version.

On Tue, Jan 29, 2019 at 02:39:49PM +0200, Sakari Ailus wrote:

quoted

quoted

+static int csi_notify_complete(struct v4l2_async_notifier *notifier)
+{
+	struct sun4i_csi *csi = container_of(notifier, struct sun4i_csi,
+					     notifier);
+	int ret;
+
+	ret = v4l2_device_register_subdev_nodes(&csi->v4l);
+	if (ret < 0)
+		return ret;
+
+	ret = sun4i_csi_v4l2_register(csi);
+	if (ret < 0)
+		return ret;
+
+	return media_create_pad_link(&csi->src_subdev->entity, csi->src_pad,
+				     &csi->vdev.entity, 0,
+				     MEDIA_LNK_FL_ENABLED |
+				     MEDIA_LNK_FL_IMMUTABLE);

This appears to create a link directly from the sensor entity to the video
device entity. Is that intentional? I'd expect to see a CSI-2 receiver
sub-device as well, which I don't see being created by the driver.

This is indeed a novel proposal. I have some concerns though.

The user doesn't have access to the configured media bus format (reflecting
the format on the CSI-2 bus on receiver's side). It's thus difficult to
figure out whether the V4L2 pixel format configured on the video node
matches what the sensor outputs. Admittedly, we don't have a perfect
solution to that whenever the DMA hardware supports multiple V4L2 pixel
formats on a single media bus format. We might need to have a different
solution for this one, should it be without that receiver sub-device.

Could you add the CSI-2 receiver sub-device, please?

Even though the name of the controller is *very* confusing, this isn't
a MIPI-CSI receiver, but a parallel one that supports RGB and BT656
buses.

Right.

quoted

quoted

+	csi->pad.flags = MEDIA_PAD_FL_SINK | MEDIA_PAD_FL_MUST_CONNECT;

Could you make it IMMUTABLE and ENABLED? If there is no need to disable it,
that is.

The link is already created with those flags, and as far as I know it
doesn't exist for the pads

Oops. And it wasn't even late when I reviewed the patch. :-P Please ignore
the comment --- I agree on the pad flags and the link flags (some lines up)
appear fine, too.

quoted

quoted

+static int csi_release(struct file *file)
+{
+	struct sun4i_csi *csi = video_drvdata(file);
+	int ret;
+
+	mutex_lock(&csi->lock);
+
+	ret = v4l2_fh_release(file);

v4l2_fh_release() always returns 0. I guess it could be changed to return
void. The reason it has int is that it could be used as the release
callback as such.

quoted

+	v4l2_pipeline_pm_use(&csi->vdev.entity, 0);
+	pm_runtime_put(csi->dev);
+
+	mutex_unlock(&csi->lock);
+
+	return ret;
+}

Do you want me to change the construct then?

Please.

-- 
Kind regards,

Sakari Ailus
sakari.ailus@linux.intel.com